Why Everyone Pays ARM?（原声文稿）

Every time a phone is made, anywhere on earth, a little bit of money flies to one company in England.

It's not the factory, it's not the brand on the back.

It's a company that for 35 years never made a single chip.

Apple pays it, Samsung pays it, Google, Amazon, Nvidia.

Companies that fight about everything all send money to the same place.

About a thousand chips a second on almost everyone a fee. The company is ARM.

It started in an old turkey barn, and its first chip was so frugal it once ran with its power cable unplugged. So why can't anyone stop paying it?

Here's the strange part. Everyone pays arm because arm promised it would never be their rival. This year it became one. Cambridge, England, 1983.

A small computer company called Acorn has a hit, the BBC Micro, the computer in British classrooms. The BBC expected it to sell 12 ,000.

It sold around 1 .5 million.

Now, Acorn needs a faster brain for its next machine. It tests the big American chips. None of them is good enough. So, Acorn asks Intel for a favor.

Herman Hauser, Acorn's co -founder, remembers the answer. And they said, get lost. Well, you get lost. We'll do our own.

There was once more problem. Nobody at Acorn had ever designed a processor. Two engineers take the job.

Sophie Wilson, who designs the instructions the chip will understand, and Steve Ferber, who designs the machine that runs them.

They fly to Phoenix, Arizona, to visit the company behind their favorite chip.

They expect a big American lab. They find a bungalow in the suburbs.

Some of the chip work is being done by local school kids in their summer holidays.

If they can design the microprocessor, than maybe we can too.

Back in Cambridge, their boss gives them two advantages. The first advantage was I gave them no people. And the second advantage was I gave them no money.

No money meant one thing. The chip had to be simple.

The big chips of the day worked like a master chef. One genius who cooks a complicated recipe all by himself. Impressive! But he needs a huge kitchen and a lot of energy.

Wilson and Ferber built an assembly line instead. Lots of tiny, simple steps. Each one fast, each one cheap.

Engineers call it risk, a reduced instruction set. The numbers tell the story.

Intel's new chip in 1985, the 386, had about 275 ,000 transistors. The tiny switches inside a chip.

Theirs had about 25 ,000.

One -tenth. Before building it, Ferber wrote a model of the whole processor to test it.

In BBC Basic, the language of the classroom computer, about 800 lines.

They called it the Acorn risk machine, ARM. There was one more rule.

The chip had to be cheap to package. Powerful chips sat in ceramic packages that could take the heat. Acorn could only afford plastic.

And plastic could take about one watt, so Ferber played it extra safe.

Effectively, we used Victorian engineering margins on the power consumption of the chip. April 26th, 1985.

The first chips arrived from the factory. By the afternoon, they're running.

Then they look at the meter on the power supply. It reads zero.

The board had a fault. The main power line wasn't connected at all. The processor was actually running on leakage from the logic circuits.

aiming for under one watt they had landed under a tenth of a watt. Wilson called the low power a complete accident.

Ferber says it was lucky but not an accident.

Either way, nobody was thinking about phones. Phones like yours didn't exist yet.

Quick pause. So far we've thrown a lot at you.

Risk. 25 ,000 transistors. vectors, Victorian margins, a meter that read zero. And we're only in 1985.

The arm went into Acorn's new computer, the Archimedes, but Acorn was in trouble.

The Italian company Olivetti had taken control, and according to Ferber, nobody had even told Olivetti about the arm when it bought Acorn.

Wilson and her colleagues had big plans for the chip.

Plans Acorn couldn't afford. So Hauser went looking for partners for three years.

The answer, he remembers, was always the same. Arm? A British company microprocessal? Nah.

The best low -power chip of its time had one customer, its owner, and its owner was running out of money. Then, a call from California. Apple.

Apple is building the Newton, a computer you hold in one hand. It needs a chip that's fast but barely sips the battery. The arm is perfect.

There's one catch. Apple won't build its future on a chip owned by a competitor.

order. So, in November 1990, the chip leaves Acorn.

A new company is born. Advanced risk machines.

Arm. Apple brings money and owns about 43%.

VLSI, the American firm that made the first chips, brings design tools. Acorn brings 12 engineers. And the headquarters?

An old turkey barn outside Cambridge. The new company needed a boss.

Apple wouldn't hand over its money until it had one.

They found Robin Saxby, a manager from the chip giant Motorola. Saxby looked at what he had. Twelve engineers. A barn.

Not nearly enough money to build a chip factory. So he made a rule.

We'll make chips over my dead body. Arm would sell the design. Others would make the chips. Think of a songwriter.

She writes the song once. Other artists record it, sell it, it, tour with it.

And every time a copy is sold, the songwriter gets a small cut. Arm wrote the song. Chip companies paid a fee up front to use it.

And then, a royalty, a small cut, on every chip they shipped, for years. That rule hit a second promise.

If Arm never made chips, it would never compete with the companies buying its designs. Everyone's partner, nobody's rival. Then, the Newton flopped.

If Arm had been a chip maker with one big customer, that could have been the end, but ARM could sell its song to anyone.

In 1993, ARM signed Texas Instruments and TI had a very big customer, a Finnish company called Nokia.

Nokia liked the chip, but ARM's instructions were long and memory was expensive.

More memory meant a more expensive phone, so ARM squeezed them.

A second, shorter set of instructions called Thumb, like packing of 1997, Nokia launched the 6110, the first mass -market phone built on that ARM chip.

And the first Nokia with a game you might remember, Snake.

Ten years later, ARM's partners were shipping nearly three billion chips a year. That's almost two in every mobile phone. And Apple? It had come close to collapse.

In 1997 alone, it lost more than a billion In 1998, Arm went public.

As Apple recovered, it started selling its arm shares.

Over the next five years, those sales brought Apple gains of about $800 million.

In one year, they were about a third of Apple's entire profit.

And in the same quarter, Apple booked one of those arm gains, it booked another payment, a $90 million bonus for Steve Jobs.

In the form of a private jet, Apple would come back to Arm in a much bigger way.

In 2007, Steve Jobs walked on stage with the first iPhone. Its brain was built on an ARM design.

The next year came the first Android phone, ARM again.

Today, ARM says more than 99 % of the world's smartphones run on its designs.

When ARM was founded, a Mac newsletter noted that Apple had promised to use only Motorola's chips in the Mac.

30 years later, Apple moved the Mac to ARM too. So why doesn't anyone just switch?

because ARM isn't really selling a chip, it's selling a language.

The architecture is essentially a common language for software developers. Every app on your phone speaks that language.

More than 22 million developers write software for ARM, the company says.

Switching would be like asking a whole country to change its language overnight.

The apps, the tools, the operating systems, all of it. And writing your own language from scratch?

ARM's own prospectus says that And in the past decade, no company has successfully designed a modern processor from zero. So you pay. How much do you pay? Very little.

By ARM's own estimate, for every dollar of chips built on its designs, ARM keeps about 1 .7 cents. But the chips never stop.

ARM's partners have shipped more than 350 billion of them.

In early 2025, that was close to a thousand every second.

earned, and selling the same song again costs ARM almost nothing.

Last year, for every dollar ARM earned, its direct costs were less than 3 cents.

Apple has signed a deal with ARM that runs beyond 2040. A business like that attracts buyers.

July 2016, 25 days after Britain voted to leave the European Union, the Japanese group Softbank agreed to buy ARM for about $32 billion. dollars.

Its founder, Masayoshi Son, had flown to a Turkish seaside resort to meet arms chairman on holiday. I did not make the investment because of Brexit.

Four years later, SoftBank agreed to sell arm to Nvidia for $40 billion. Arm's customers were alarmed. Nvidia competes with many of them.

Whoever owns arm could see their plans and decide who gets what and when.

America's Union opened deep investigations.

Chip companies, the watchdog said, saw arm as the Switzerland of chips, neutral.

I think it's an absolute disaster for Cambridge, the UK, and Europe. February 2022, the deal collapsed.

That same day, Arm got a new boss, René Haas.

He had spent seven years at Nvidia. September 2023, Arm returns to the stock market.

And look Look who lined up to buy shares. Apple, Google, Nvidia, Samsung, Intel, Customers, Rivals, all investing in the same toll booth.

But ARM had a problem. Almost everyone who wants a smartphone already has one.

How do you grow when the number of phones doesn't? You raise the price of the song.

ARM's newer designs, called ARM V9, carry at least double the royalty of the ones before.

and ARM started selling bigger, ready -made blocks of chip at even higher rates.

In early 2025, by analysts' estimates, phone shipments grew less than 2%.

ARM's royalties from smartphones grew about 30%. Same phones, bigger bill.

And according to company documents shown in court and reported by Reuters, ARM had once discussed raising some prices by as much as 300%.

One customer decided it had paid enough.

Qualcomm, the company behind the Snapdragon chips in many Android frames.

In 2021, Qualcomm bought a startup called Nuvia, which had its own ARM license.

Later, in court, Qualcomm CEO told the jury that internal documents projected the deal could save Qualcomm up to $1 .4 billion a year in payments to ARM.

ARM sued. It said Nuvia's technology had to be destroyed.

December 2024. A courtroom in Delaware.

Arm CEO Rene Haas is on the stand, and he says this, We don't build chips. The jury sided with Qualcomm.

15 months later. March 24th, 2026. Arm unveils the Arm AGI CPU.

A data center chip with up to 136 cores made by TSMC, designed by ARM, sold by ARM. The first chip of its own.

Its lead partner is Meta, a customer Qualcomm had also been trying to win.

Today marks the next phase of the ARM compute platform and a defining moment for our company. Why now?

Data centers are running out of power, and the giants want efficient chips, fast.

Arm says demand for its chip already tops $2 billion, and it is aiming for a $15 billion business.

And according to Reuters, Arm had started recruiting executives from its own customers for this back in November 2024.

Weeks before its CEO told a jury, we don't build chips. Arm's own annual report now spells it out.

Many of our semiconductor and systems company customers who have historically licensed our IP may face direct competition from us.

Saxby's rule had lasted 35 years. October 2026.

The same courthouse in Delaware. This time, Qualcomm is suing ARM.

Qualcomm's entire chip business depends on ARM and on ARM honoring its contract and honoring its promises. They were not harmed in the least.

October 13th. The jury decides. ARM broke its contract by holding back test material it owed Qualcomm.

And it interfered with Qualcomm's business with Meta.

On the form, the jury had to say what kind of relationship the two companies have.

It circled two. One of them was competitor. The damages? One dollar. And one more dollar.

Arm won the arguments about pricing, and said the award of only $2 proved its point. But Qualcomm isn't after the $2.

It is asking the judge to let it stopped paying ARM royalties for up to five years. Qualcomm is about 9 % of ARM's revenue.

As we record this, the judge hasn't decided.

And Qualcomm is packing a bag.

It bought a company that designs chips with Risk 5, a free, open alternative to ARM's language. No license, no royalty.

According to notes from the courtroom, a Qualcomm plan shows Risk 5 phone chips by 2029.

Meta, Arm's first chip customer, reportedly bought a Risk 5 startup too.

Nvidia already puts Risk 5 inside its own chips, about a billion of those cores in 2024 alone.

Even Intel and AMD, rivals for decades, joined forces to defend their own chip language. None of them has left Arm. Risk 5 is still young.

The judge herself reportedly questioned whether it is a real alternative yet. But the money behind ARM is under pressure too.

SoftBank has pledged 72 % of ARM's shares to borrow money, reportedly for its AI bets, and ARM's stock is worth roughly half what it was in June. So why does everyone pay ARM?

Because two engineers with no money built a chip so simple it ran on almost nothing.

Because a company in a barn couldn't afford a factory, so it sold the song instead, to everyone.

And because it promised to stay neutral, rival, everyone's partner, nobody's rival.

That promise is why Apple, Samsung, Google, and Nvidia kept paying, even when they hated the bill.

Now ARM sells its own chips, it is raising the toll.

And one of its biggest customers has just shown a jury that ARM broke its word, even if it only cost $2.

Robin Saxby used to finish his rule with a joke, I'm still alive and ARM doesn't make chips. Today, only half of that joke is still true.

The next time you pick up your phone – cool, quiet, no fan – picture that little coin flying to Cambridge.

It all started with a meter that read zero, and a turkey barn that turned it into a toll on almost every chip on Earth. So here's our question.

If you ran Apple, or Qualcomm, what would you do? Keep paying the toll, or start learning a new language? Tell us which side you'd pick, down in the comments.